Ever find yourself in a collaborative situation that’s gone awry? Maybe one person on the team dominates the conversation or an offhand comment distracts everyone for a whole meeting. Experts in social computing are developing a virtual mediator to turn around unproductive discussions.
"The basic problem is that there is a lot of untapped potential of working as a group," said Yla Tausczik, a University of Austin at Texas graduate who worked on the new system as part of her dissertation with psychology department chair James Pennebaker. Building on previous work in this area, the two researchers designed feedback system that gives real instruction instead of just nudges.
As a teaching assistant for introductory psychology at the university, Tausczik watched students struggle. During a discussion about narcissistic personality disorder or NRD, one student sidetracked the group by saying things like "my roommate totes has NRD" and "my roommate is so effing narcissistic." Students ended up talking about what they all know rather than adding new and relevant information.
Other researchers have been trying to make systems to get groups back on track for a while, Tausczik pointed out. A social visualization tool developed at Cornell University called GroupMeter was designed for an online collaboration. It tracks the words written while people work on a task together via chat and shows their level of agreement.
Pennebaker and Tausczik, who's now a postdoctoral fellow at Carnegie Mellon University’s Human-Computer Interaction Institute, wanted to make a more powerful tool. Instead of simply making a group more aware of how they are working together, the researchers created a real-time language feedback system that actively monitors online discussions and gives recommendations for the group to follow.
First, students in the class logged into an online educational platform they use to hold 20-minute group chats for their psychology class. While they did that, the researchers' newly developed computer system ran in the background. Although this system doesn't have a name yet, it was recently presented at the Conference on Human Factors in Computing Systems in Paris.
The system starts by tracking the words each member of a group uses. When the students in a group use similar words in categories that are considered psychologically meaningful, that usually indicates the participants are paying attention to each other. Too much overlap can mean they're just rehashing the material, though. The system then processed the typed discussions, calculated scores, and generated suggestions at regular intervals that appeared in pop-up boxes for each group.
If a group wasn't using enough similar words early on, the system recommended the participants pay more attention to each other. Then if the system detected that the discussion had improved, it praised the group with an encouraging pop-up saying they were doing a good job.
Results were mixed, though.
When the recommendations were too harsh or didn't fit with the students' perception of how they were doing, the students hated it. Critiques read "Yo this pop-up sucks" and "death to the green box." But Tausczik said that when the feedback was accurate, the students appreciated it. One wrote, "Our group got back on task when the reminder popped up."
The researchers learned that the ideal system should be selective with feedback. Next, they plan to expand the system so it takes into account the more nuanced ways that groups collaborate. This might mean making more role-dependent recommendations. A student with a better grasp on the material might be encouraged to play a Socratic role and challenge her classmates' ideas.
Beyond the classroom, Tausczik imagines the system being used in conflict management, work settings, and even email clients. She suggested that a virtual mediator could generate an alert if the email message you’re composing is getting too off-topic. It would work sort of like the Gmail's forgotten attachment reminder.
"I see it as a tool that someone would want to use because it helps them get better perspective on the way they interact with others," Tausczik said. "The interaction could go smoother."
Gilly Leshed is a visiting assistant professor in Cornell University's department of information science who specializes in human-computer interaction and worked on GroupMeter. While GroupMeter was tested with superficial tasks in a lab setting, Tausczik and Pennebaker's system was deployed in a real classroom.
"They went beyond the piece of the raw data," Leshed said. "They processed it for the group and gave them a prescription of how to respond." She added that she liked the idea of a virtual mediator for email.
"Imagine an email that can show you the degree to which you are being too critical or you're being too dominant or too subversive," she said. "Today with technologies like Siri you can do it even in voice communication."
Speech-to-text capabilities are constantly improving, Leshed added. With the right computer system, collaboration could be a breeze.